Ollama v0.31.2-rc2 Enables iGPU Multimodal Projector Offload

Core summary

One sentence to understand this update

Ollama v0.31.2-rc2 introduces support for offloading multimodal projector (mmproj) operations to integrated GPUs (iGPUs) on non-Metal systems by using fit padding.

Impact & opportunity

What this could mean

Developers using Ollama can now leverage iGPUs for multimodal models on a wider range of hardware, potentially improving performance and reducing reliance on discrete GPUs.